EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Adaptive Load Balancing in Distributed Web Search Caches via Hybrid Selective Query Replication and Dynamic Consistent Hashing Ranges

Carlos Gómez-Pantoja, Alonso Inostrosa-Psijas, Verónica Gil-Costa, Mauricio Marin

IEEE Access · 2026 · 14 · 90135-90152

DOI: 10.1109/access.2026.3702877 ↗

Kurz gefasst

Fachartikel in der Zeitschrift IEEE Access, 2026. Verfasst von Carlos Gómez-Pantoja und 3 weiteren Autor:innen. Die Arbeit nennt 56 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: Fondecyt de Iniciación from Agencia Nacional de InvestigaciÃn y Desarrollo (ANID), Chile, Sciences et Technologies de l’Information et de la Communication (STIC)-AmSud Program.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Distributed Cache Services are critical components in Web Search Engines, designed to reduce query latency and alleviate pressure on indexing subsystems. However, the efficiency of these services is frequently compromised by the highly skewed Zipfian distribution of user queries. This phenomenon creates “Structural Imbalance,” where a small subset of frequent queries overloads specific nodes regardless of the hashing strategy used. Furthermore, “Dynamic Imbalance” arises from temporal shifts in query popularity and stochastic fluctuations in node availability. Moreover, Distributed Cache Services operate in highly dynamic environments where processing nodes are frequently added or removed due to scaling operations, failures, and fault-tolerance mechanisms. These topology changes induce abrupt workload reassignments, further exacerbating load heterogeneity and transient overloads. As a result, infrastructure volatility compounds both structural and dynamic imbalance, posing additional challenges for maintaining stable performance. This paper presents a comprehensive, dual-layer solution to these inefficiencies. First, we propose a “long-term” balancing strategy that uses the Space-Saving algorithm to identify frequent queries and distribute their processing across multiple replica nodes, effectively flattening structural skew. Second, we introduce a “medium-term,” data-dependent dynamic load-balancing mechanism based on the Receiver-Initiated Diffusion algorithm, which adjusts the range responsibilities of nodes within a Consistent Hashing ring in real time. Extensive simulations using a 133-million-query trace segment drawn from a representative sampled log collection of a commercial Web Search Engine demonstrate that, in the single-layer evaluation, selective replication (PRB) reduces average response time by 18.3% relative to standard Consistent Hashing. In the dual-layer evaluation, our dual-layer (PRDL) strategy achieves a workload efficiency of 88.6% in failure-free operation and 86.8% under 16.7% simultaneous node failures, while maintaining response time below 30 ms in both regimes.

Förderung

  • Fondecyt de Iniciación from Agencia Nacional de InvestigaciÃn y Desarrollo (ANID), Chile
  • Sciences et Technologies de l’Information et de la Communication (STIC)-AmSud Program

Zitieren

APA

Gómez-Pantoja, C., Inostrosa-Psijas, A., Gil-Costa, V., & Marin, M. (2026). Adaptive Load Balancing in Distributed Web Search Caches via Hybrid Selective Query Replication and Dynamic Consistent Hashing Ranges. IEEE Access, 14, 90135-90152. https://doi.org/10.1109/access.2026.3702877

Vancouver

Gómez-Pantoja C, Inostrosa-Psijas A, Gil-Costa V, Marin M. Adaptive Load Balancing in Distributed Web Search Caches via Hybrid Selective Query Replication and Dynamic Consistent Hashing Ranges. IEEE Access. 2026;14:90135-90152. doi:10.1109/access.2026.3702877

BibTeX
@article{GmezPantoja2026,
  author = {Gómez-Pantoja, Carlos and Inostrosa-Psijas, Alonso and Gil-Costa, Verónica and Marin, Mauricio},
  title = {Adaptive Load Balancing in Distributed Web Search Caches via Hybrid Selective Query Replication and Dynamic Consistent Hashing Ranges},
  journal = {IEEE Access},
  year = {2026},
  volume = {14},
  pages = {90135--90152},
  doi = {10.1109/access.2026.3702877}
}
Kostenlos starten →